What you’ll be able to do
Five options, no clear winner, and everyone online arguing about which one is “best” like it’s a football team. It isn’t. It’s just software, and picking one is easier than the internet makes it look.
By the end of this you’ll have picked one AI assistant from the five main options (ChatGPT, Claude, Gemini, Copilot, Perplexity), run it through a proper test using your own real work, and know whether paying for it would actually get you anything. Not vibes, not a YouTuber’s opinion. Your own three tests, on your own tasks.
That’s genuinely it. Takes about twenty minutes.
Before you start
Right, quick tour before we touch anything.
ChatGPT and Claude are general chat assistants, good for writing, explaining, planning, coding, the everyday stuff. Gemini is Google’s version of the same thing, and on a paid Google AI plan it works inside Gmail and Docs too, which matters if you’re already a Google person. Copilot is Microsoft’s, and it makes the most sense if you’re already using Word, Excel or Outlook at work. Perplexity is different: it’s search-first, meaning every answer comes with the sources it actually read, so it’s the one to reach for when you want to click through and check something rather than just trust it.
There’s also NotebookLM, which is a bit of a sleeper option. It only answers from documents you upload yourself, with citations, which makes it brilliant for “read these forty pages and tell me what matters” jobs. And if your data genuinely can’t leave your machine, ever, there’s running something locally with Ollama or LM Studio, free and private, though noticeably less capable. That’s its own guide though, “How to get AI to run on my own computer”, so I won’t get into it here.
Here’s the decision rule I’d actually use: start where you already are. Microsoft 365 at work? Copilot. Gmail, Android phone, or Google Workspace? Gemini. Neither of those? ChatGPT or Claude, pick either. Research job with sources to check? Perplexity. Your own documents? NotebookLM. Data that can’t leave the building? Local.
And here’s the honest bit, the bit most comparison articles won’t tell you: for everyday stuff like emails, summaries, explanations, plans, the top four are close. Properly close. A free tier is enough to find out which one suits you. You switch later when you hit an actual wall (a message limit, a feature you need like Projects), not because someone online said their model is smarter this week.
So. Start free. Always start free.
Get set up
Pick one from the list above based on where you already work. Then:
ChatGPT: go to chatgpt.com, sign up with an email or a Google account, you’re in. Free plan gives you unlimited everyday chats on a smaller model, though file uploads, image generation, voice chat, deep research and memory are all limited on free.
Claude: claude.ai, same deal, quick signup. Free includes chat on web, desktop and mobile, web search, file creation, running code, memory across conversations, connecting apps, voice mode, and you get the Sonnet and Haiku models. You can even make up to five Projects on the free plan, which is more generous than people expect.
Gemini: gemini.google.com, and if you’ve got a Google account you’re basically already set up. Free gets you the Flash model with some Pro access, image creation and editing, Deep Research, Gemini Live for voice, Canvas, Gems, and 15 GB of storage shared across Gmail, Drive and Photos. Worth knowing: memory and personalisation need a personal Google account, not a work or school one.
Copilot: copilot.microsoft.com, free with a Microsoft account, works on web, in Windows, and on mobile. The Microsoft 365 Personal, Family and Premium subscriptions put Copilot inside Word, Excel, PowerPoint, Outlook and OneNote with more usage than the free app; Microsoft 365 Basic doesn’t include the AI features, and on Family only the subscription owner gets them. The old separate Copilot Pro plan is gone; its page now redirects to Microsoft 365 Premium.
Perplexity: perplexity.ai, free to use, no faff.
Once you’re in, go find the privacy settings. This matters more than people think. By default, the consumer plans of ChatGPT, Claude and Gemini may use your chats to help train their models unless you opt out. OpenAI’s own help centre says consumer plans “may be used for training depending on whether you have opted out.” Anthropic’s privacy policy says the same sort of thing, your chats are used “unless you opt out through your account settings.” Google has a “Keep Activity” switch at myactivity.google.com/product/gemini, and it’s worth knowing that chats reviewed by humans can be kept for up to three years. Find the toggle, decide what you’re comfortable with, move on.
I haven’t verified how Copilot or Perplexity handle this, so I’m not going to guess. Check their own privacy pages if it matters to you.
Try it yourself
Right, this is the actual test. Open two of the assistants side by side (browser tabs work fine) and run the exact same three prompts on each.
First, ask it to help you decide in the first place, just to see how it reasons about itself:
I'm choosing an AI assistant. Here's what I'll mostly use it for: [list]. I already use [Microsoft 365 / Google Workspace / neither]. Which one should I start with, and why, in five lines?Then the real tests. An email reply:
Reply to this email in my tone. Under 120 words, no bullet points: [paste the email]A summary job:
Summarise this article in five bullet points and tell me the one number that matters: [paste]And explaining something you actually know:
Explain [something you know well] to a complete beginner in 150 words.Use a real email you’ve actually got sitting in your inbox. Use an actual article you were going to read anyway. And for the last one, pick something you genuinely know, your job, a hobby, whatever, because you’re the only one who can judge if the explanation is any good.
Check the result
Now compare. Which one’s email reply sounds like something you’d actually send, versus something you’d need to rewrite? Which summary captured the one number that mattered, and which one buried it or missed it? Which explanation would actually work on a beginner, and which one used jargon it was supposed to be avoiding?
Pick the one whose answers you’d edit least. That’s basically the whole test. Not which one sounds cleverest, not which one has the flashiest features, just which one gets closest to usable on the first go.
Then use it for a week, properly, for real tasks, and pay attention to where it stops you. Did you hit a message limit halfway through a busy Tuesday? Did you need to upload a file and get told no? Did you want a feature like Projects or scheduled tasks and find it’s paid-only? Write those down as they happen, don’t try to remember them later, you won’t.
At the end of the week you’ll know one of three things: stick with the free plan because it’s doing the job, switch to a different assistant because this one keeps annoying you in the same spot, or pay because you now know exactly what feature you’re paying for and it’s worth it. All three are fine outcomes. The point was never to end up on the “best” one, it was to end up on the one that works for you.
If it doesn’t work
Few common snags.
Hit a message limit? Wait for it to reset, use a second assistant for the rest of the day, or pay if it’s happening every single day, not just once.
Need your own files properly answered, not summarised badly? Try NotebookLM, since it only answers from what you upload, with citations, or step up to a paid tier on your main assistant.
Got an answer that felt off, or just wrong? Check the sources yourself. Better still, ask Perplexity that specific question next time, since it shows you what it read.
And the bit that actually matters: don’t hand any of these tools information you wouldn’t want stored somewhere you don’t control. Passwords, banking details, anything medical or legal that identifies a real person, confidential work documents under NDA, other people’s personal information. None of that goes into a chat box. Doesn’t matter how good the privacy settings look, treat every chat as something that could, in theory, be read by a human reviewer somewhere down the line, because for some plans, it might be.
All of this, by the way, is available in the UK, no workarounds needed.
Questions people ask
Which AI assistant should I choose?
Start where you already are. Microsoft 365 at work? Copilot. Gmail, Android or Google Workspace? Gemini. Neither? ChatGPT or Claude, pick either. Research job with sources to check? Perplexity. Your own documents? NotebookLM. Data that can't leave the building? Local.
Can I get a good AI assistant for free?
Yes, always start free. For everyday stuff like emails, summaries, explanations and plans, the top four are properly close, and a free tier is enough to find out which one suits you. Switch or pay later when you hit an actual wall.
How do I test ChatGPT vs Claude on my own work?
Open two assistants side by side and run the exact same three prompts on each: an email reply in your tone, a five-bullet summary with the one number that matters, and an explanation of something you know well to a beginner. Use a real email and a real article.
How do I know which assistant gave the better answer?
Pick the one whose answers you'd edit least. Not which sounds cleverest or has the flashiest features, just which gets closest to usable on the first go. Then use it for a week and note where it stops you.
What should I never type into an AI chat?
Don't hand these tools information you wouldn't want stored somewhere you don't control: passwords, banking details, anything medical or legal identifying a real person, confidential work documents under NDA, or other people's personal information.
Keep exploring
Once you’ve picked something and used it for a bit, two things tend to come up. First, you’ll notice your prompts aren’t getting you quite what you want, which is where “How to get AI to understand what I want” comes in. Second, if privacy is bugging you more than the opt-out toggles can fix, “How to get AI to run on my own computer” walks through running something locally instead. And if you’re annoyed that it forgets everything between conversations, “How to get AI to remember context” covers that properly.
Sources and review notes
https://openai.com/chatgpt/pricing/ https://claude.com/pricing https://gemini.google/subscriptions/ https://www.microsoft.com/en-gb/microsoft-365/buy/compare-all-microsoft-365-products https://www.perplexity.ai/ https://support.google.com/gemininotebook/answer/16213268 https://openai.com/policies/privacy-policy/ https://www.anthropic.com/legal/privacy https://support.google.com/gemini/answer/13594961 https://myactivity.google.com/product/gemini
Review date: 22 September 2026